Académique Documents
Professionnel Documents
Culture Documents
IMF Working Papers describe research in progress by the author(s) and are published to elicit comments
and to encourage debate. The views expressed in IMF Working Papers are those of the author(s) and
do not necessarily represent the views of the IMF, its Executive Board, or IMF management.
© 2017 International Monetary Fund WP/17/283
Abstract
The impact of real exchange rate movements on GDP growth is a hotly debated issue
both in policy and academic circles. In this paper we provide evidence suggesting that the
association between exchange rate misalignment and growth for a broad panel of countries is
very weak. Controlling for country …xed e¤ects, time e¤ects and initial GDP, a more depre-
ciated currency is associated with higher growth if one does not exclude outliers. However,
this positive association always vanishes after controling for the savings rate. Importantly,
this applies for both a large panel of countries and for the emerging economies subsample.
The views expressed herein should be attributed to the authors and not to the IMF, its Executive Board, or
its management.
y
A¢ liation: IMF and University of Sao Paulo, cgoncalves@imf.org
z
Associate Professor, Department of Economics, University of Sao Paulo, Brazil. Email: mrodrigues@usp.br.
1
1 Introduction
In this paper we revisit the exchange rate-growth nexus using panel data estimations. Our
…ndings do not lend support to the generally accepted view that a more depreciated exchange
Even if one assumes the real exchange rate can be used as a policy lever, deviations from
its fundamental value on growth rates is ambiguous. In a model with learning by doing/scale
economies, a depreciated currency can boost growth by temporarily shielding domestic produc-
ers from import competition, thus allowing them to travel up the production learning curve
where long term productivity prospects are more promising, it could spark higher overall growth
rates. On the other hand, if imported capital goods play a signi…cant role in domestic capital
formation, making them dearer through an undervalued currency may cause a fall in investment
It is a priori hard to say which e¤ect should prevail. Rodrik (2008) is a major reference in
the empirical literature dealing with this question. Using data from 1950 to 2004 from Penn
World Table (PWT 6.2), he …nds that a measure of exchange rate misalignment is associated
with GDP per capita growth rates: undervalued currencies leading to higher growth rates, and
overvalued currencies to lower growth rates. The author claims this e¤ect is due to currency
misalignments in emerging economies (he …nds no signi…cant e¤ect for advanced economies).
Here we use the same methodology, but employ data from the latest version of the Penn
World Table (PWT 9.0). For a broad panel of countries, we show that the relationship between
real exchange rate deviations and growth either is not there, or vanishes after controlling for an
2
important omitted variable: the savings rate.
In a world with imperfect capital markets, internal savings may matter for growth. For
instance, in Aghion et al (2016) model, domestic savings increase the attractiveness of FDI
which boosts growth1 . At the same time, though, higher savings are associated with smaller
current account de…cits and thus with a less depreciated exchange rate. This implies a positive
association between misalignment and growth may simply re‡ect these two variables’correlation
with savings.
There is large body of research investigating the growth e¤ects of exchange rates. Some
papers emphasize the deleterious e¤ects of overvaluation (Sachs and Warner, 1995; Razin and
Collins, 1997; Easterly, 2005; Rajan and Subramanian, 2011). Rodrik (2008) …nds evidence
that not only overvaluation is bad for growth, but also that an undervalued currency tends to
spur growth. Using a large panel of countries, he proceeds in two stages. In the …rst stage, he
regresses real exchange rate against per capita income and time …xed e¤ects. Residuals from this
regression are then taken as a measure of undervaluation of the local currency. In the second,
he regresses GDP per capita growth against the undervaluation variable and other controls.
Here we follow the same methodology, but use more recent data (1950 to 2014). Furthermore,
we emphasize the role of savings as a possible confounder in the estimation of the undervaluation-
growth relationship.
Berg and Miao (2010), who control for other possible determinants of the real exchange
rate in the …rst stage, …nd evidence that …rst-stage residuals are highly correlated with GDP
per capita growth. Gluzman, Levy-Yeyati and Sturzenegger (2012) suggest the growth e¤ects
1
The logic is that foreign ownership is better protected from expropriation if the company has a domestic
partner: for political economy reasons, the sovereign is less likely to expropriate domestic players, attenuating the
agency problem and rendering FDI ‡ows safer.
3
of an undervalued currency work through savings, investment and employment. Levy-Yeyati,
Sturzenegger and Gluzman (2013) …nd evidence that countries intervene to avoid overvaluation,
and these interventions are related to faster growth. Again, these e¤ects are particularly strong
full panel (which includes developed and developing countries), the e¤ect of undervaluation
on growth is positive, but becomes statistically insigni…cant and its magnitude is substantially
reduced once we add savings as a control. Moreover, we do not …nd evidence that the e¤ect
when we exclude outrageous outliers, the estimated e¤ect of exchange rate deviations turns
The rest of the paper is organized as follows: section 2 lays down the methodology used,
For our benchmark estimations we rely on data from the latest version of the Penn World Table
(PWT 9.0), which encompasses the period between 1950 and 2014. We consider 13 …ve-year
periods: 1950-54, 1954-60, ..., 2010-14. Unless otherwise noted, variables refer to averages across
each period.
Estimation proceeds in two stages. In the …rst stage we regress the real exchange rate2 on a
set of time dummies and on GDP per capita levels. As in Rodrik (2008), the idea is to control
2
The real exchange rate is equal to the nominal exchange rate (local currency/USD) divided by the local price
level (PPP). The U.S. price level is normalized to 1.
4
for Balassa-Samuelson type of e¤ects and global macro shocks3 .
where RERit is country i’s average real exchange rate over period t, Yit is country i’s average
real GDP per capita over period t, and t are time …xed e¤ects. We then compute a measure of
deviations/misalignments, which are the residuals of the equation above. We call this variable
Dev_RERit .
The second stage is a stripped-down version of a Barro growth regression with exchange rate
where git is the average annual growth rate of GDP per capita in period t, country i; Yit0 is the
initial GDP per capita of country i, period t;4 i and t are, respectively, country and time
…xed e¤ects. We include ln Yit0 to capture standard convergence e¤ects. Moreover, Savingsit
is country i’s average savings rate over period t, de…ned as one minus the shares of private
consumption and government consumption in GDP. We run regressions with and without the
Savings are a possible confounding factor in estimating the exchange rate-growth relation-
ship. Higher savings can both boost growth and contribute to a more devalued currency at the
same time. As shown below, savings and exchange rate deviations are indeed strongly correlated.
3
This di¤ers from IMF’s misalignment measure, which employs the e¤ective exchange rate instead.
4
Yit0 is real GDP per capita of country i in the …rst year of period t.
5
Furthermore, the inclusion of savings reduces substantially the magnitude and precision of the
3 Estimations
N = 1899; R2 = 0:456
where values in parenthesis are robust t-statistics (the regression includes a full set of time
dummies). Our results are very similar to Rodrik’s. GDP per capita is negatively associated with
the log-levels of the exchange rate, meaning more developed economies tend to have stronger
currencies. This is the famous Balassa-Samuelson e¤ect: higher productivity in the tradable
sector increases wages in both tradable and non-tradable sectors, which in turn put a pressure
on non-tradable prices leading to a stronger currency. The statistical signi…cance of this variable,
Table 1 displays descriptive statistics on our main variables. We show separate statistics for
emerging economies, which we de…ne as countries with GDP per capita smaller than 70% of
that of the US and larger than USD 1,000 (data from very poor countries are in general less
reliable).
6
Table 1: Descriptive statistics
We have argued above that savings and real exchange rate deviations are likely positively
correlated. Therefore, the e¤ect of undervaluation on growth would be biased upward if savings
were excluded in exchange rate-growth regressions. In our sample, this correlation is indeed pos-
itive and highly signi…cant. Table 2 reports the outputs of panel regressions of RER deviations
against savings (full sets of country and time …xed included). We present results for our full
sample of countries and for the subsample of emerging economies. In both cases the coe¢ cient
7
Table 2: Correlation between RER deviations and savings
(4.84) (3.82)
N 1899 1401
We next report second-stage results (see Table 3). The dependent variable is the average annual
growth rate multiplied by 100. Column (1) exhibits results for the whole sample, which combines
both developed and developing countries. In addition to the deviation variable, we include the
log of initial GDP per capita plus time and country …xed e¤ects. Using this speci…cation, we
…nd a signi…cant e¤ect of exchange rate deviations on growth (see column (1)). The magnitude
is large, although somewhat smaller than the one reported by Rodrik (in a similar regression his
coe¢ cient of Dev_RER is around 1.7, compared to 1.3 found here). A one standard deviation
increase in this variable is associated to a growth rate higher by 1:318 0:454 0:60 percentage
points. This is a big number, given mean growth rate in the sample is 2.3%.
However – and this is the main message from this paper – when we include savings in
our regression (column (2)), the estimated e¤ect of Dev_RER becomes insigni…cant, with the
8
magnitude of the point estimate being cut in half. It is important to stress that insigni…cancy
is not due to in‡ated standard-errors. It is the point estimate that drops by a lot.
We repeat the analysis in column (3) and (4), now removing outliers.5 This time, the
coe¢ cient of RER deviation is insigni…cant even when we do not control for savings. Moreover,
the e¤ect of savings is large and very precisely estimated. For each percentage point increase
Columns (5)-(8) are analogous to columns (1)-(4), but restrict the sample to emerging
economies only. Results are broadly similar. Importantly, we do not …nd stronger growth
e¤ects of devaluations for this group of countries. In some speci…cations (as in columns (5) and
(6)), the estimated coe¢ cients of Dev_RER are actually smaller than those found for the full
sample. In any case, when we control for savings, the e¤ect of devaluations becomes insigni…cant
once again. This stands in sharp contrast with Rodrik’s …nding that devaluations have stronger
5
We remove, as outliers, country-period pairs that displayed: (i) average GDP per capita growth rates larger
than 10 or smaller that -5; or (ii) average devaluations rates smaller than -5 or larger than 5; or (iii) average
savings rates smaller than zero or larger than 1.
9
Table 3: Second stage growth regressions 6
Dependent variable: Average annual growth rate of GDP per capita (multiplied by 100)
10
Savingsit 3:028 10:21 3:224 9:447
6
Emerging Economies are those with GDP/head smaller than 70% of that of the US and higher than USD 1,000.
3.3 Overvaluations versus Undervaluations
Linearity may be a strong assumption: overvaluations can be fundamentally di¤erent from un-
dervaluations in terms of growth impacts –as mentioned, many papers investigate the deleterious
e¤ects of RER overvaluations on growth. To assess the importance of this kind of asymmetric
where Pit is equal to 1 if Dev_RERit 0 (i.e., currency is undervalued), and equal to zero
Now the marginal impact of an overvalued currency is captured by coe¢ cient , whereas
the impact of an undervalued currency is given by + : Table 4 displays the results of this
exercise. We only show regressions that include savings as a control variable. For our full
sample of countries, the coe¢ cients of both Dev_RER and the interaction variable are statis-
tically insigni…cant (columns (1) and (2) below). In other words, we …nd neither evidence that
undervaluations are good for growth, nor that overvaluations are bad growth.
Columns (3) and (4) report the same estimations for our sample of emerging economies. In
column (3) we actually see a positive impact of overvaluations on growth ( > 0) and negative
impact of undervaluations on growth ( + < 0). However, these coe¢ cients may capture the
fact that crisis periods in emerging markets are both characterized by low growth and capital
‡ight (giving rise to undervaluations); and in large expansions we should observe the opposite.
Indeed, when we exclude outliers (in particular, very large overvaluations and undervaluations),
the coe¢ cients of Dev_RERit and Pit Dev_RERit become once again insigni…cant (see
11
column (4)).
Dependent variable: Average annual growth rate of GDP per capita (multiplied by 100)
12
4 Concluding remarks
In this short paper we investigate the impact of real exchange rate misalignments on growth. We
follow closely Rodrik (2008), which …nds evidence that a depreciated real exchange rate tends to
boost growth, based on a panel of countries. His results are particularly strong for a subsample
of developing economies.
Here we use a more recent version of PWT data, while emphasizing the importance of
controlling for domestic savings. The exclusion of this variable may bias upward the e¤ect of
the real exchange rate, since higher savings can both increase growth and contribute to a more
depreciated currency. Indeed, in our sample, RER deviations and savings are strongly correlated.
For a broad panel of countries, we do …nd that RER deviations are positively associated
with growth. However, this e¤ect becomes statistically insigni…cant when we include savings as
a regressor. This also holds true for a subsample of emerging economies. We also estimate an
alternative speci…cation which allows for di¤erent e¤ects of undervaluations and overvaluations
To summarize, contrarily to Rodrik’s famous result, we do not …nd evidence that RER
misalingments matter for growth, either for a broad set of countries or for a subsample of
emerging economies.
References
[1] Aghion, P.; Comin, D.; Howitt, P.; Tecu, I. 2016. “When do domestic savings matter for
13
[2] Berg, A.; Miao, Y. 2010. “The Real Exchange Rate and Growth Revisited; The Washington
Consensus Strikes Back?” IMF Working Paper #10/58, International Monetary Fund.
[3] Easterly, W. 2005. “National Policies and Economic Growth: A Reappraisal.” In: Aghion,
P.; Durlauf, S. (ed.). Handbook of Economic Growth, volume 1, chapter 15, p. 1015-1059.
[4] Gluzmann, P. A.; Levy-Yeyati, E.; Sturzenegger, F. 2012. “Exchange rate undervaluation
and economic growth: Díaz Alejandro (1965) revisited.”Economics Letters 117(3): 666-672.
[6] Rajan, R. G.; Subramanian, A. 2011. “Aid, Dutch disease, and manufacturing growth.”
[7] Razin, O.; Collins, S. M. 1997. “Real Exchange Rate Misalignments and Growth.” NBER
[8] Rodrik, D. 2008. “The Real Exchange Rate and Economic Growth.” Brookings Papers on
[9] Sachs, J. D.; Warner, A. 1995. “Economic Reform and the Process of Global Integration.”
14